Research Proposal - Engineering Informatics Group - Stanford ...

CIFE Center for Integrated Facility Engineering •Stanford University
CIFE Seed Proposal Summary Page
2005-2006 Projects
Proposal Title: Computational Modeling of Nonadaptive Crowd Behaviors
for Egress Analysis
Principal Investigator(s): Kincho H Law, Jean-Claude Latombe,
and Ken Dauber.
Research Staff: Xiaoshan Pan
Proposal Number: (Assigned by CIFE): 200501
Total Funds Requested: $69,191
First Submission?
No If extension, project URL:
http://eil.stanford.edu/egress/
Abstract (up to 150 words):
Safe egress is one of the key design issues identified by facility planners, managers and
inspectors. Current computational tools for the simulation and design of emergency
egress rely heavily on assumptions about human individual and social behaviors, which
have been found to be oversimplified, inconsistent and even incorrect. This research
aims to develop a framework for modeling human and social behaviors from the
perspectives of human decision-making and social interaction and to incorporate such
behaviors in a dynamic computational model suitable for emergency egress analysis.
For the year-1 project, we have developed a theoretical framework of crowd behavior
and a proof-of-concept, multi-agent based crowd simulation model. In the year-2 project,
we plan to study the scalability and modularity issues of the simulation framework, to
validate the models, and to incorporate engineering analyses, such as performancebased
assessment of facilities, design of egress and emergency plans.
Law, Latombe, Dauber Egress Simulation 1

Computational Modeling of Nonadaptive Crowd
Behaviors for Egress Analysis
1. ENGINEERING PROBLEM
The main objective of this research is to develop a computational framework that can
facilitate the study of human and social behavior for emergency exits in buildings and
facilities. Design of egress for places of public assembly is a formidable problem in facility
and safety engineering. Although provisions governing egress design are prescribed in
code, the actual performance of evacuation systems is difficult to assess. There have
been numerous incidents reported regarding overcrowding and crushing during
emergency situations. In addition to injuries and loss of lives, the accompanying postdisaster
psychological suffering, financial loss, and adverse publicity have long-term
negative effects on the individuals and organizations – the survivors, the victims’ families,
the corporations involved and the communities [1].
In a crowded environment, it has been observed that most victims were injured or killed
by the so called “nonadaptive” behaviors of the crowd, rather than the actual cause (such
as fire or explosion) of the disaster. Nonadaptive crowd behaviors refer to the destructive
actions that a crowd may experience in emergency situations, such as stampede,
pushing, knocking, and trampling on others, etc.; these actions are responsible for a large
number of injuries and deaths in man-made and natural disasters. For example, in the
Iroquois Theatre fire (in 1903), the initial fire was brought under control quickly; however,
602 people were trampled to death in the end. Another example is the Hillsborough
English FA Cup Stampede (in 1981); there were no real causes of emergency but still 95
people died and over 400 people were injured. To study nonadaptive behavior in a
crowded environment, we need to take into consideration the human and social behavior
in emergency situation from both psychological and sociological perspectives.
Building codes contain “means of egress” provisions designed to ensure the safety of a
building [2]. However, these codes only provide basic guidelines and are not exhaustive
and often insufficient for many practical situations [7]. First, current codes and guidelines
contain inconsistencies which may lead to misinterpretations. An effective computational
tool can test whether a specific guideline is appropriate for a particular situation. Second,
each building is unique, and compliance with design guidelines does not automatically
ensure safety. Often, local geometries – shapes and sizes of spaces and obstacles –
can have significant influence to egress, albeit in a subtle way. To date, very few studies
can be found in existing literature in terms of understanding how environmental
constraints and local geometries impact crowd behaviors and movements. This type of
studies is difficult since it often requires exposing real people to the actual, possibly
dangerous, environment. A good computational tool which takes into consideration of
human and social behavior of a crowd could serve as a viable alternative. Computational
tools are now commercially available for the simulation and design of emergency
evacuation and egress. However, these tools focus on the modeling of spaces and
occupancies and aim mostly to animate crowd movements but they rarely take into
consideration human and crowd behaviors.
This research proposes to study human behaviors and social interactions, and to
incorporate such behaviors in a dynamic computational model suitable for safe egress
Law, Latombe, Dauber Egress Simulation 2

analysis. The computational model will be able to adapt egress analysis suitable for
specific design circumstances and provide insight to current prescriptive and often,
ambiguous codes and provisions for egress design. Also, the model can serve as a
means to study safety engineering, such as assessing building codes and designs,
testing safety and evacuation procedures, and crowd management.
For the year-1 seed project, we have so far developed a theoretical framework of crowd
behavior and built a proof-of-concept crowd simulation model. In the year-2 project, we
plan to study the scalability and modularity issues of the simulation framework, to
validate the models, and to incorporate engineering analyses, such as performancebased
assessment of facilities, design of egress and emergency plans.
2. POINTS OF DEPARTURE
The proposed research will build upon the current understanding of nonadaptive crowd
behavior in emergencies. Generally speaking, existing theories on crowd behavior in
emergency situation can be classified into three basic categories: (1) panic [9-12], (2)
decision-making [13,14], and (3) urgency levels [15]:
• Panic theories deal primarily with the factors that may cause panic during
emergencies. The basic premise is that when people perceive danger, their usual
conscious personalities are often replaced by the unconscious personalities which in
turn lead them to act irrationally unless there is a presence of a strong positive social
(such as a leader) influence.
• Decision-making theories assume that a person, even under dangerous situation, can
still make (albeit limited) rational decisions, attempting to achieve good outcomes and
objectives in the situation [14]. These “rational” decisions however may or may not
improve the overall emergency evacuation. In a situation such as a fire, cooperating
with others and waiting one’s own turn can likely be beneficial to the group and, in
turn, increasing the individual’s likelihood of exiting a facility. On the other hand, if
some people are pushing, then an individual may feel threatened if he/she does not
react; the best course of action for the individual may be to join the competition and
push, in order to maximize his/her chance of exiting.
• Another theory suggests that the occurrence of (human) blockages of exiting space
depends on the levels of urgency to exit [15]. There are three crucial factors that
could lead to such situation: the severity of the penalty and consequence for not
exiting quickly, the time available to exit, and the group size. A problem arises when
the urgency to leave reaches a high level of anxiety – for example, too many people
try to exit quickly at the same time. Thus, any efforts that can reduce the number of
people having a high urgency to leave will likely cause a decrease in jams and less
entrapment.
There have been a wide variety of computational tools, many of them are now
commercially available, for egress simulation and design of exits. Most existing models
can be categorized into fluid or particle systems, matrix-based systems, and emergent
systems:
• Many have considered the analogy between fluid and particle motions (including
interactions) and crowd movement. Two typical examples of fluid or particle systems
are the Exodus system [4] and the panic simulation system built by Helbing et al. [6].
Coupling fluid dynamic and “self-driven” particle models with discrete virtual reality
simulation techniques, these systems attempt to simulate and to help design
evacuation strategies. However, as noted by Still [7], “the laws of crowd dynamics
Law, Latombe, Dauber Egress Simulation 3

have to include the fact that people do not follow the laws of physics; they have a
choice in their direction, have no conservation of momentum and can stop and start at
will.” Fluid or particle analogies also contradict with some observed crowd behaviors,
such as herding behavior, multi-directional flow, and uneven crowd density
distribution.
• The basic idea of a matrix-based system is to discretize a floor area into cells. Cells
are used to represent free floor areas, obstacles, areas occupied by individuals or a
group of people, or regions with other environmental attributes. People transit from
cell to cell based on occupancy rules defined for the cells. Two well known examples
of the matrix-based systems are Egress [3] and Pedroute [5], which have been
applied to simulate evacuation in buildings as well as train (and underground)
stations. It has been revealed that existing matrix-based models cannot simulate
crowd cross flow and concourses; furthermore, the assumptions employed in these
models are questionable when compared with field observations [7].
• The concept of emergent systems is that the interactions among simple parts can
simulate complex phenomena such as crowd dynamics [16]. One example of the
emergent systems is the Legion system [7]. It should be noted that Legion was not
designed as a crowd behavioral analysis system but an investigation tool for the study
of large scale interactive systems. Current emergent systems typically oversimplify
the behavioral representation of individuals. For example, the Legion system employs
only four parameters (goal point, speed, distance from others, and reaction time) and
one decision rule (based on assumption of the least effort) to represent the complex
nature of individual behaviors. All individuals are considered to be the same in terms
of size, mobility, and decision-making process. Finally, the model ignores many
important social behaviors such as herding and leader influence.
In summary, as noted by the Society of Fire Protection Engineers [8], “These
[computational] models are attractive because they seem to more accurately simulate
evacuations. However … they tend to rely heavily on assumptions and it is not possible to
gauge with confidence their predictive accuracy,” and “the fundamental understanding of
the sociological and psychological components of pedestrian and evacuation behaviors is
left wanting [17].”
3. PROPOSED RESEARCH
3.1 Research Methods
Figure 1 depicts the basic iterative approach undertaken to study the human and social
behaviors and to develop a computational framework for egress simulation and analysis.
The research starts with literature studies and field interviews on crowd behaviors in the
areas of human decision-making, social interaction, safety engineering, and computer
simulation. We then derive the variables, hypotheses, and human behavior rules and
patterns, which are then formalized and incorporated into a multi-agent based simulation
framework. The results are in turn analyzed to capture the patterns of crowd behavior.
The research results are evaluated and validated through comparing with prior
observations and historical cases as well as by practicing experts in the field. Observed
and derived information are then incorporated to the framework for further simulation.
Law, Latombe, Dauber Egress Simulation 4

Literature review,
Data collection
Extract variables
and rules
Observed behavior,
Hypotheses
Simulation
model
Analysis of results,
“pattern extraction”
Validation,
Theory development
Integration
Figure 1: Research process
Geom. Engine
Population
Generator
Physical
Geometry
Individual
Agents
Crowd Simulation Engine
Individual
Behavior
Model-1
Individual
Behavior
Model-4
Individual
Behavior
Model-2
. . .
Individual
Behavior
Model-3
Individual
Behavior
Model-n
Events
Agents’
Positions
Events Recorder
Visualizer
Global Database
(Global geometry information and the
state information of the agents)
Figure 2: System Architecture
The proposed system architecture of the simulation framework is schematically shown in
Figure 2. The current system is based on a multi-agent simulation framework which
consists of five basic components: a Geometric Engine, a Population Generator, a Global
Database, a Crowd Simulation Engine, an Events Recorder, and a Visualization
Environment.
• Geometric Engine: The purpose of this module is to produce the geometries
representing the physical environments (e.g., a building or a train station, etc.).
AutoCAD/ADT (Architectural Desktop Software from AutoDesk, Inc.) is employed in
this study. The geometric data is sent to the Crowd Simulation Engine to simulate
crowd behaviors.
• Population Generator. This module generates virtual agents to represent a crowd
based on the distribution of age, mobility, physical size, and type of facility (hospital,
office building, train station, stadium, etc.) to be investigated. The population and its
composition for each type of facility would be different. For example, we can assume
most (not all) of the occupants in an office building will likely be familiar with the
facility; on the other hand, the same assumption cannot be applied to a theme park.
This module also generates random populations for statistical study of individual
human behaviors and crowd behaviors.
• The Global Database. The database module is to maintain all the information about
the physical environment and the agents during the simulation. Although the multiagent
system does not have a centralized system control mechanism, the state
information (mental tension, behavior level, location) of the individuals are
maintained. This database is also needed to support the interactions and reactions
among the individuals.
Law, Latombe, Dauber Egress Simulation 5

• The Events Recorder. This module is intended to capture the events that have been
simulated for retrieval and playback. The events captured can be used to compare
with known and archived scenarios for evaluation purpose.
• The Visualizer. The visualization tool is to display the simulated results. We have
developed a simple visualization environment that is able to receive the positions of
the agents, and then generates and displays 2D/3D visual images.
• The Crowd Simulation Engine. This module is the core module of the multi-agent
simulation system. Each agent is assigned with an “individual behavior model” based
on the data generated from population generator. The Individual Behavior Model is
designed to represent an individual human’s decision-making process. Each agent
responds to its environment using the decision rules and initiate actions and reactions
accordingly. The internal mechanism of the Individual Behavior Model consists of the
following iterative steps: (1) internally trigger for decision; (2) perceive information
about the situation (i.e., crowd density, sensory input, tension level); (3) interpret
and choose decision rule(s) to make a decision; (4) conduct collision check and
execute the decision. Each autonomous agent will proceed to the (exit) goal
subjected to the constraints imposed, acquire the behavior rules, select appropriate
actions, interact with and update the Global Database as simulations proceed over
time.
In addition to the simulation of crowd behaviors, the outputs of the system will also
include overall and individual evacuation time, individual paths, and blockage locations.
The simulation system is designed with sufficient modularity to allow further investigation
of crowd dynamics and incorporation of new behavior patterns and rules as they are
discovered.
3.2 Summary of Research Accomplishments and Further Investigations
For the first year seed project, we have studied a number of fundamental issues and built
a prototype system. The following briefly summarize our accomplishments so far in this
project.
• Thorough review of human and social behavior literatures in the field of psychology,
sociology, safety engineering, and egress simulation. These literatures have
confirmed the originality of the proposed work and supported the fundamental
behavioral theories that are being investigated. Furthermore, a series of interviews
with emergency personnel (fire marshals, police personnel, faculty members in social
science, police science at Stanford University and other campuses). We have also
evaluated a number of egress simulation tools (such as Exodus, Simulex, Egress,
Pedroute, Legion and others) and “behavior animation” tools (such as Massive, Koga,
etc..). The review so far has revealed the lack of human and social behavior in
current simulation and animation systems.
• Establishing a theoretical framework to study human and social behavior in
emergencies. The framework dissects the study of complex human and social
behavior into three interdependent levels: individual, interaction among individuals,
and group. Some fundamental factors and behavior pattern of crowd behavior are
identified and formalized for constructing a simulation framework. This theoretical
framework will also allow us to further investigate the dynamic changes of behavior
during an emergency situation.
• Developed a proof-of-concept prototype to simulation some frequently observed
crowd behaviors. A multi-agent based prototype has been built and is currently able
to demonstrate some typical crowd behaviors such as competitive, queuing, herding
behaviors and bi-directional crowd flow (as illustrated in Figure 3). Incorporating more
Law, Latombe, Dauber Egress Simulation 6

ehaviors including some fatal overcrowding conditions are currently underway.
Shortcomings of the current computational models have also been identified.
Competitive behavior
Queuing behavior
Herding behavior
Bi-directional crowd flow
Figure 3: Some behavioral patterns demonstrated using the prototype.
The research results so far have been accepted for publication and presentation at a
number of academic and building engineering conferences [18-20]. These publications
and the simulation results can be found in the project website:
http://eil.stanford.edu/egress.
Results from current prototype study are indeed encouraging, judging from the interviews
with the safety personnel and comments received during the advisory meetings.
However, much research efforts, ranging from the study of human and social behaviors to
the development of simulation framework, remain. In addition to continuing the work
described in the first year funding proposal, the second year seed project will include
fundamental research study on the following critical issues:
• So far, we have studied and incorporated some selected behaviors (competitive,
herding, queuing, etc..). However, the so called “non-adaptive” behaviors, which are
the causes of the injuries and fatalities, have not been formalized and systematically
studied. This continuing effort will develop models that will include into consideration
of pushing and other psychological factors (such as audible sounds).
• While we have demonstrated the prototype system successfully with a population of
over 400 agents, the scalability of the system will be a key issue that needs to be
studied carefully. Fast algorithms developed in the AI and Robotics program of the
CS department will be adopted to improve the sensory capabilities and the crowd
Law, Latombe, Dauber Egress Simulation 7

movements. We envision that these fast algorithms can provide a scalable platform
to handle thousands of agents for crowd simulation.
• We envision that the prototype system will be extended and used to integrated with
other engineering analyses, such as performance-based assessments of facilities and
spaces, design of egress and safety/emergency plans. The framework is currently
being re-designed to provide sufficient modularity for such integration.
3.3 Research Impact
In safety engineering, because few efforts have been conducted to study the core of
crowd safety problem – nonadaptive crowd behavior – from human psychological and
sociological perspectives, the proposed research is expected to fulfill this gap and will
subsequently lead to significant contributions to the field of crowd safety research, which,
due to recent natural and man-made events, is fast becoming an important issue in
facility design. A simulation tool such as the one proposed can be used to help facility
design, emergency exits, evacuation plan. In addition, these types of simulation can help
develop a wider range of possible solution to crowd problems. The tool can provide
valuable information to guide the planning process, for construction, to develop crowd
management strategies and for dealing with emergencies.
3.4 Milestones and Anticipated Risks
We anticipate that by the end of the first year period, the basic issues related to “nonadaptive
crowd behaviors” will be thoroughly reviewed and the prototype system will be
re-designed for modularity and scalability. For the proposed second year research period
from September 2005 through August 2006, the research milestones are set up as
follows:
• By the end of Fall Quarter 2005 – we plan to complete the preliminary implementation
of a scalable simulation system, which allows (1) simulations for a significant number
of agents in the crowd, and (2) predefined deterministic or random assignments of
individuals in design spaces.
• By the end of Winter Quarter 2006 – we plan to test and to complete the
implementation so that we can compare the simulated results and replicate the
patterns of some historical overcrowding incidents and case studies on some real-life
crowd settings (e.g., sport events and subway stations).
• By the end of Spring Quarter 2006 – we plan to analyze our findings and, if
necessary, to develop a tool that will enable the analysis of large scale simulated
results which may involve hundreds of randomized simulations for particular settings.
• By the end of Summer Quarter 2006 – we plan to complete a detailed report on this
study. Furthermore, we anticipate that we will produce a range of experimental
results by applying the simulation system to assess the performance of a variety of
real facilities.
This research is a high-risk, high-payoff, interdisciplinary project (with participants from
Civil Engineering, Computer Science and Sociology) that could lead to significant
advancement in facility engineering and design. We anticipate the following risks:
• Human and social behaviors are complex subjects. The crowd behavior model that
we aim to develop may not be sufficiently general for a broad range of scenarios. We
address this risk by design the system with sufficient flexibility and modularity to allow
further investigation of crowd dynamics and incorporation of new behaviors as they
are discovered.
Law, Latombe, Dauber Egress Simulation 8

• The simulation of large number of crowds may require a significant amount of
computations. We anticipate that a number of fast algorithms developed by the AI and
Robotics group will be incorporated to ensure system scalability. If necessary, the
PIs have significant experience in distributed and parallel computing research, and
we do expect available computational platforms available to the PIs are sufficient to
ensure the usability of the simulation systems.
4. RELATIONSHIP TO CIFE GOALS
Safe egress is one of the most critical issues in facility engineering and design. This
research supports the “function” thrust area of this year’s Call For Proposal through
developing a framework to help “assess how well the occupied facility meets the initial
and then-current functional objectives” from safety engineering perspective. This
research is highly interdisciplinary and can lead to immediate industrial applications,
which closely match the CIFE mission.
5. INDUSTRY INVOLVEMENT
This project has received significant feedbacks from industries (e.g., Sydney Olympic
Park, IR Security & Safety, and Disney) and other agencies. We will work with any
organizations who have interests to be involved in our continuing research effort.
6. NEXT STEPS
Research proposal is currently under preparation for the new Human and Social
Dynamics Program initiative from NSF. Furthermore, the PI is a member from Stanford in
a pre-proposal for an NSF’s Engineering Research Center for Fire Engineering and
Safety. We will continue to pursue government funding opportunities such as NSF, NIST,
FEMA, Office of Homeland Security and others. This research towards the development
of simulation system for emergency exits and egress design could have impact on
emergency responses due to natural and man-made disasters. Successful demonstration
of this research may lead to practical products needed by the industry.
REFERENCES:
[1] Lystad, M., Mental Health Response to Mass Emergencies: Theory and Practice,
Brunner/Mazel, New York, 1988.
[2] ICBO, “Means of Egress”, 2000 International Building Code, Chapter 10, pp. 211-
247, 2000.
[3] AEA Technology, A Technical Summary of the AEA EGRESS Code, Technical
Report, AEAT/NOIL/27812001/002(R), Issue 1, 2002, available at http://www.aeatsafety-and-risk.com/Downloads/Egress%20Technical%20Summary.pdf.
[4] Fire Safety Engineering Group, BuildingEXODUS: the Evacuation Model for the
Building Environment, Sept. 2003, available at
http://fseg.gre.ac.uk/exodus/air.html#build.
[5] Halcrow Group Limited, Pedroute, Sept. 2003, available at
http://www.halcrow.com/pdf/ urban_reg/pedrt_broch.pdf.
[6] Helbing, D., Farkas, I., and Vicsek, T., “Simulating Dynamical Features of Escape
Panic,” Nature, 407:487-490, 2000.
[7] Still, G., Crowd Dynamics, Ph.D. thesis, University of Warwick, UK, 2000.
[8] Society of Fire Protection Engineers, Engineering Guide to Human Behavior in Fire,
Technical Report, 2002, available at
http://www.sfpe.org/sfpe/pdfsanddocs/DraftHumanBehaviorGuide.pdf.
[9] La Piere, R., Collective Behavior, McGraw-Hill, New York, 1938.
Law, Latombe, Dauber Egress Simulation 9

[10] Le Bon, G., The Crowd, Viking, New York, 1960. (Original publication, 1895)
[11] McDougall, W., The Group Mind, Cambridge University Press, Cambridge, UK,
1920.
[12] Smelser, N., Theory of Collective Behavior, Free Press, New York, 1963.
[13] Brown, R., Social Psychology, Free Press, New York, 1965.
[14] Mintz, A., “No-Adaptive Group Behavior,” Journal of Abnormal and Social
Psychology, 46: 150-159, 1951.
[15] Kelley, H., Condry, J., Dahlke, A, and Hill, A., “Collective Behavior in a Simulated
Panic Situation,” Journal of Experimental Social Psychology, 1:20-54, 1965.
[16] Epstein, J., and Axtell, R., Growing Artificial Societies: Social Science from the
Bottom Up, MIT Press, Cambridge, MA, 1996.
[17] Galea, E., (Ed.), Pedestrian and Evacuation Dynamics, Proceedings of 2 nd
International Conference on Pedestrian and Evacuation Dynamics, London, UK,
CMC Press, 2003.
[18] Pan, X., Han, C., Dauber, K. and Law, K. "Human and Social Behavior in
Computational Modeling and Analysis of Egress," BFC Doctoral Program, Las
Vegas, March 16 - 17, 2005.
[19] Pan, X., Han, C. and Law, K. "A Multi-agent Based Simulation Framework for the
Study of Human and Social Behavior in Egress Analysis," The International
Conference on Computing in Civil Engineering, Cancun, Mexico, July 12 - 15, 2005.
[20] Pan, X., Han, C., Dauber, K. and Law, K. "A Multi-agent Based Framework for the
Simulation of Human and Social Behaviors during Emergency Evacuations," Social
Intelligence Design 2005, Stanford University, Stanford, USA, March 24 - 26, 2005.
Law, Latombe, Dauber Egress Simulation 10

Project: CEE-FY05-290 --LAW: CIFE PROPOSAL BUDGET FOR 2005-2006
Department: Civil Engineering
Principal Investigator: LAW, KINCHO (Prof) - CE
Administrator: L. Unerdem
Period 1 All Periods
10/01/05 - 9/30/2006 10/01/05 - 09/30/06
% Amount Total Amount
Salaries
Senior Personnel
Law, Kincho (Prof) acad 0 0 0
smmr 16 6,823 6,823
Latombe, J. (Prof) acad 0 0 0
smmr 8 4,477 4,477
Dauber, Kenneth (Assoc Dir) cal 3 3,787 3,787
Graduate Students
PhD Student, CE (Res Asst) acad 50 21,154 21,154
smmr 50 7,052 7,052
Total Salaries 43,293 43,293
Benefits
Faculty 4,783 4,783
Graduate 959 959
Total Salaries and Benefits 49,035 49,035
Travel, Domestic
Domestic Travel 1,500 1,500
Costs Not Subject to IDC Costs
Tuition
Tuition 18,656 18,656
Total Direct Costs 69,191 69,191
Modified Total Direct Costs 50,535 50,535
University IDC Costs
Annual Amount Requested 69,191 69,191
Rates Used in Budget Calculations
Benefit Rates
Faculty: UFY06 31.70%; UFY07 31.70%;
Graduate: UFY06 03.40%; UFY07 03.40%;
Indirect Cost Rate
Special Rate: UFY06 00.00%; UFY07 00.00%;
Law, Latombe, Dauber Egress Simulation 11